Feature request: optional cleanup of source folder after confirmed successful hardlink/copy

[JBendarr]

As discussed on #1027 - a proper writeup for this rather than a PR, since I'm not a dev myself.

Feature request: optional cleanup of source folder after confirmed successful hardlink/copy

Situation: a download client (e.g. qBittorrent) saves directly into the same root a library scanner (e.g. Audiobookshelf) watches. This is a valid setup, and one the compose examples half-suggest by mounting the client's path directly. Shelfmark's current behaviour here is correct by design - is_managed_workspace_path() in shelfmark/download/postprocess/workspace.py only ever treats TMP_DIR as mutable, and says why:

# workspace.py:50-54
def is_managed_workspace_path(path: Path) -> bool:
    """Return whether Shelfmark should treat this path as mutable.
    The managed workspace is `TMP_DIR`. Anything outside it should be treated as
    read-only for safety (e.g. torrent seeding directories).
    """

That's the right default. But the original torrent-delivered folder survives every successful hardlink or copy, and the library scanner then registers it as a second item alongside the real one. Hit this myself with a couple of test audiobooks - harmless with hardlinking (same inode, no extra disk used), but it leaves duplicate library entries every time, and I imagine it'd be worse for anyone using copy mode (HARDLINK_TORRENTS_AUDIOBOOK off), where it'd double disk usage per download too.

Ask: an opt-in, off-by-default setting to clean up the source folder, only after positively confirming the transfer succeeded.

• Hardlinked output: confirm the destination shares the source's inode before deleting.
• Copied output: confirm matching size (optionally a hash) before deleting.
• Exclude anything still actively seeding by default. Gate on the torrent completion action being remove, or a separate seed-time/ratio guard, so this never fights someone who wants to keep seeding.

Separate, smaller thing while I'm here: the current Remove completion action doesn't delete files for torrents at all - delete_files=False is hardcoded for the torrent branch in base_handler.py, against delete_files=True for the equivalent usenet path. Nothing in the settings copy or environment-variables.md signals that distinction, and it's easy to assume otherwise (I did). Might be a cheap docs/UI fix even before the bigger feature above.

Scope note: doesn't apply to the ebook pipeline the same way, since CWA already does import-then-delete on the destination side. This is specifically about library tools with no equivalent watch-then-consume step.

Understand you're focused on bugfixes for now so no rush on this, just wanted to get it written up properly. Thanks for taking the time to look at #1027.